Polymarket 5-min market-making · paper

One row per market — are we participating, and are we winning? Participation: quoted (we made a market), or the reason we sat out — near-the-money / push (the reversal signal telling us to withdraw before the manipulators hit) or no two-sided book. Total vol = all taker shares this cycle (both tokens); two-sided = the matchable part (gold). Net = spread captured − inventory settled at resolution ($0 if the wrong side won). Volume in shares; toggle UTC⇄IST up top. Paper only, no orders placed.

Net = spread captured (the steady, repeatable edge) ± inventory (leftover shares settled at the true up/down outcome — can help or hurt, and it's high-variance). Inventory is booked only when we know the real settlement; 1019 cycles still hold unrealized inventory, so treat net as provisional. The trustworthy line is the spread; market-making only truly pays if net clears the inventory swings.

Bars = cycles per hour · orange = near-the-money cycles (the contested ones ② withdraws from). Two label rows: UTC hour and IST (= UTC + 5:30). Manipulation clusters in the thin-liquidity Asia/overnight window (≈ 21:00–07:00 UTC = 02:30–12:30 IST) — where making a market is riskiest.

Every taker trade on both the Up and Down tokens, folded onto one probability axis (buy Down ≡ sell Up). Green up = toward-Up pressure, red down = toward-Down, at each 1¢. A gold tick marks levels with genuine two-sided flow — the price bands where a maker could actually round-trip (rest a buy below, a sell above, and have both fill). If the gold ticks cluster somewhere (e.g. 40–60¢), that's the makeable band; if it's all green-at-99¢ / red-at-1¢ with nothing two-sided in the middle, market-making has nothing to work with here.